Machine Learning Weight Update via Evaluation Value Calculation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning using Boltzmann machines requires a large number of product-sum operations to update weights, leading to increased learning time, which is inefficient and power-intensive, especially in resource-constrained environments.
Innovation Solution
A machine learning system with a learning section and an operating section that updates weights by calculating an evaluation value based on input data and internal state, reducing the number of product-sum operations necessary for weight updates and optimizing learning time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If Boltzmann machine is used for machine learning, then the number of times of updating weights necessary for learning can be reduced, but the number of times of product-sum operation necessary to update weights once increases
Solution Approach 1:
The patent divides the weight update process into two distinct phases: (1) a precomputation phase where intermediate values are calculated and stored, and (2) a weight update phase where the actual weight updates occur. This segmentation allows the system to perform fewer weight updates while maintaining accuracy, thereby reducing learning time despite the increased complexity of each individual update operation.
Solution Approach 2:
The patent performs preliminary calculations of intermediate values (such as activation values and their derivatives) before the actual weight update process. By precomputing and storing these values, the system avoids redundant calculations during the weight update phase, which reduces the overall number of operations needed and accelerates the learning process.
2Productivity
If the number of times of updating weights is reduced, then learning time is shortened, but the number of times of product-sum operation increases
Solution Approach 1:
The patent changes the computational parameters by introducing intermediate representation values that capture essential features of the data. By transforming the input data into these intermediate representations and performing operations on them, the system reduces the number of full product-sum operations needed, thereby lowering computational energy consumption while maintaining or improving learning speed.
3Loss of time
If Boltzmann machine learning is applied, then learning can be completed in shorter time, but power consumption increases due to increased product-sum operations
Solution Approach 1:
The patent extracts and separates the computationally intensive product-sum operations from the weight update process. By identifying and isolating these operations, the system can optimize them independently or eliminate redundancies, thereby reducing overall power consumption while maintaining the accelerated learning timeline provided by the Boltzmann machine approach.
Data Source
AI summary
A machine learning system includes a learning section and an operating section including a memory. The operating section holds a required accuracy, and an internal state and a weight value of a learner in the memory and executes calculation processing by using data input to the machine learning system and the weight value held in the memory to update the internal state. An accuracy of the internal state is calculated from a result of the calculation processing and an evaluation value is calculated using the data input to the machine learning system, the weight value, and the updated internal state held in the memory when the calculated accuracy is higher than the required accuracy. The evaluation value is transmitted to the learning section, which updates the weight value by using the evaluation value and notifies the number of times of updating the weight value to the operating section.